JP2019503105A - Multi-resolution compressed sensing image processing - Google Patents

Abstract

Systems and methods are provided for constructing a multi-resolution image from a set of compressed measurements that represent a compressed version of a single native resolution image. In one aspect, a compressed sense measurement is retrieved that represents a compressed version of a single native resolution image of a scene generated using a compressed sensing matrix. A desired two-dimensional multi-resolution image size is determined, a plurality of areas are allocated, and each allocated area has a respective resolution. An expansion matrix is defined to map the resolution of each allocated area to the native resolution, and a multi-resolution image of the scene is constructed using the compressed sense measurements, the compressed sensing matrix, and the defined expansion matrix Is done. In various aspects, a full resolution or other multi-resolution image can be constructed from the same compressed measurements without regenerating new compressed measurements.

Description

  The present disclosure is directed to systems and methods for compressed sense image processing. More particularly, the present disclosure is directed to constructing a multi-resolution image from a compressed measurement that represents a single resolution image.

  This section provides aspects that may help to enable a better understanding of the systems and methods disclosed herein. Accordingly, the text in this section should be read from this perspective and should not be understood or interpreted as an admission as to what is prior art or not.

  Digital image / video cameras acquire and process a significant amount of raw data that is reduced using compression. In conventional cameras, raw data for each of the N-pixel images representing the scene is first captured and then generally compressed using a compression algorithm suitable for storage and / or transmission. Although compression after capturing a high resolution N-pixel image is generally useful, it requires significant computational resources and time. Furthermore, compression of the raw image data after it has been acquired does not always result in meaningful compression.

  A more recent approach, known in the art as compressive sense imaging, directly obtains compressed data for an N-pixel image (or multiple images in the case of video) of a scene. To do. Compressed sense imaging is an algorithm that uses random or sparse projection to directly generate compressed measurements for later construction of an N-pixel image of a scene without collecting conventional raw data of the image itself Implemented using Direct acquisition of a reduced number of compressed measurements compared to the more conventional method of first acquiring raw data for each of the N-pixel values of the desired N-pixel image and then compressing the raw data As such, compressive sensing significantly removes or reduces the resources required to compress an image after it has been completely acquired. An N-pixel image of the scene is constructed from compressed measurements for rendering on the display or other applications.

E. J. et al. Candes and M.C. B. Wakin, “An Induction to Compressive Sampling”, IEEE Signal Processing Magazine, Vol. 25, No. 2, March 2008 E. J. et al. Candes, “Compressive Sampling”, minutes of the International Congress of Materials, Madrid, Spain, 2006 E. Candes et al., "Robust unintention principals: Exact signal reconstitution from high infrequent information," IEEE Trans. on Information Theory, Vol. 52, No. 2, pages 489-509, February 2006

  In various aspects, systems and methods for compressed sense image processing are provided.

  One aspect is the step of retrieving M compressed sense measurements generated using an MxN compressed sensing matrix, wherein the compressed sense measurements are a single native resolution N-pixel image of the scene. Retrieving a compressed version of the image, determining a desired two-dimensional multi-resolution image size to be reconstructed from the compressed sense measurements, and determining a plurality of regions based on the determined size. Allocating to a desired two-dimensional multi-resolution image, wherein each of the allocated regions has a determined pixel resolution equal to or less than the native resolution of a single native resolution image And the expansion matrix (ex defining a size of each expansion matrix, wherein the size of each expansion matrix is determined for each respective allocated region based on the determined pixel resolution of the allocated region, and compression Generating a multi-resolution reconstructed image of the scene using the sense measurements, the compressed sensing matrix, and the defined enhancement matrix.

  One aspect includes allocating a plurality of regions after compression measurements are taken.

  One aspect includes allocating at least one region of interest having a determined pixel resolution higher than other allocated regions.

  One aspect includes reconfiguring the allocated region of interest by changing one of the allocated regions of interest or by adding additional regions of interest.

  One aspect includes generating at least one region of interest based on manual input.

  One aspect includes automatically generating at least one region of interest.

  One aspect includes displaying on the display generating a multi-resolution reconstructed image of the scene.

FIG. 7 illustrates an example compressed sense imaging system in accordance with various aspects of the present disclosure. FIG. 6 illustrates an exemplary process for reconstructing an image of an object from compressed measurements using a reconstruction basis matrix in accordance with various aspects of the present disclosure. FIG. 6 illustrates exemplary steps for allocating regions to images according to aspects of the present disclosure. FIG. 6 illustrates exemplary steps for determining resolution for an allocated region in accordance with aspects of the present disclosure. FIG. 4 illustrates exemplary steps for defining an extension matrix, in accordance with aspects of the present disclosure. FIG. 11 illustrates an example apparatus for implementing various aspects of the disclosure.

  Various aspects of the disclosure are described below with reference to the accompanying drawings, wherein like numerals refer to like elements throughout the description of the figures. The description and drawings merely illustrate the principles of the disclosure. Those skilled in the art will be able to devise various configurations that embody the principles of the present disclosure and fall within the spirit and scope of the present disclosure, although not explicitly described or illustrated herein. Will be understood.

  The term “or” as used herein is unless otherwise specified (eg, “or else” or “or in the alternative”). ), Refers to a non-exclusive OR. Further, as used herein, terms used to describe a relationship between elements should be interpreted broadly to include direct relationships or the presence of intervening elements, unless otherwise specified. It is. For example, when an element is referred to as being “connected” or “coupled” to another element, the element can be directly connected or directly coupled to another element, or there can be intervening elements. In contrast, when an element is referred to as being “directly connected” or “directly coupled” to another element, there are no intervening elements present. Similarly, terms such as “between” and “adjacent” should be interpreted in a similar manner.

  Compressed sensing, also known as compressed sampling, compressed sensing or compressive sampling, is a known data sampling technique that exhibits improved efficiency over conventional Nyquist sampling. is there. Compressed sampling allows a sparse signal to be represented and reconstructed using far fewer samples than the number of Nyquist samples. When the signal has a sparse representation, the signal can be reconstructed from a small number of measurements from a linear projection onto the appropriate basis. Furthermore, the reconstruction has a high probability of success when a random sampling matrix is used.

である）。信号ベクトルが、その信号ベクトルに線形的に関係する領域においてスパースである場合、Ｎ次元信号ベクトルは、センシング行列φを使用して、Ｍ次元圧縮された測定値ベクトルから再構築（すなわち、近似）され得る。 Compressed sensing is generally characterized mathematically as multiplying an N-dimensional signal vector by an M × N-dimensional sampling or sensing matrix φ to obtain an M-dimensional compressed measurement vector, where M is generally Much smaller than N (ie for compression,

Is). If the signal vector is sparse in the region linearly related to the signal vector, the N-dimensional signal vector is reconstructed (ie, approximated) from the M-dimensional compressed measurement vector using the sensing matrix φ. Can be done.

  Additional details regarding conventional aspects of compression sampling can be found in, for example, E.I. J. et al. Candes and M.C. B. Wakin, “An Induction to Compressive Sampling”, IEEE Signal Processing Magazine, Vol. 25, No. 2, March 2008, E.I. J. et al. Candes, “Compressive Sampling”, minutes of the International Congress of Materials, Madrid, Spain, 2006; Candes et al., "Robust unintention principals: Exact signal reconstitution from high infrequent information," IEEE Trans. on Information Theory, Vol. 52, No. 2, pages 489-509, February 2006.

In an imaging system, a compressed measurement acquired by a compressed sensing imaging device to represent a compressed version of a one-dimensional representation x (x ₁ , x ₂ , x ₃ ... X _N ) of an N-pixel image of a scene The relationship between values or samples y _k (kε [1... M]) can be expressed in matrix form as y = Ax (as shown below), where (also known as φ) A) is the MxN sampling or sensing matrix implemented by the compression imaging device to obtain the compressed sample vector y.

The vector x (x ₁ , x ₂ , x ₃ ... X _N ) is itself a one-dimensional representation of a two-dimensional (eg, row and column) native image and represents a row or column of the two-dimensional image. Known methods such as concatenating to a single column vector can be used to mathematically represent a two-dimensional image of known size as a one-dimensional vector and vice versa; Will be understood. The matrix A shown above has a maximum length sensing, or each row (also known as a basis vector) has N values corresponding to the reconstruction of the full resolution N-pixel desired image x of the scene. Also known as sampling matrix.

It should be noted that the focus of the present disclosure is not on the generation of compressed measurements y _k (kε [1... M]), where the compressed measurements are taken from conventional compressed sensing devices. It is assumed that it is given or received. However, a brief description for generating a compression measurement is given below in FIG.

Instead, the focus of the present disclosure is on the reconstruction of the desired image from the compressed samples, and in particular on the construction of multi-resolution images derived from the compressed measurements. In other words, the aspects of the present disclosure are based on compressed measurements y _k (kε [1...) Obtained using the M × N maximum length sensing matrix A to represent a compressed version of a single native resolution image. M]) for the reconstruction of the desired multi-resolution image.

  The systems and methods disclosed herein may not always be necessary in applications such as medical imaging, object recognition or security, or to generate full native resolution images from compressed measurements, or It can be advantageously used in other applications that can consume more resources than desired. In particular, the systems and methods disclosed herein are capable of generating full native resolution images from acquired compressed samples if it should be desired or necessary to do so. Do not exclude that.

  These and other aspects of the invention will now be described in more detail below with reference to the figures.

である。当業者によって理解されるように、圧縮測定値ｙ_ｋ（ｋ∈［１．．．Ｍ］）は、シーン１０４の（たとえば、２次元画像の行の連結を使用して、２次元画像の１次元表現として表される）単一ネイティブ解像度Ｎ‐ピクセル画像ｘ（ｘ_１、ｘ_２、ｘ_３．．．ｘ_Ｎ）の圧縮されたバージョンを表す。 FIG. 1 illustrates a schematic example of a compressed imaging reconstruction system 100 (“system 100”) in accordance with various aspects of the present disclosure. Incident light 102 that may be reflected from the scene 104 (which may be in the visible or invisible spectrum) is received by the compressed sensing camera device 106, which receives the compressed measurements y _k (kε [1. M]) using a predetermined maximum length sensing or sampling matrix A to generate a vector y, where

It is. As will be appreciated by those skilled in the art, the compression measurement y _k (kε [1... M]) is obtained from the scene 104 (eg, using a row concatenation of 2D images). Represents a compressed version of a single native resolution N-pixel image x (x ₁ , x ₂ , x ₃ ... X _N ) (represented as a dimensional representation).

  For example, incident light 102 reflected from scene 104 may be received at camera device 106, where the light passes through an N-element array of individually selectable aperture elements (eg, N micromirrors). Partially allowed to pass or not pass and hit the photon detector. Any or all of the N individual aperture elements at any particular time are either partially or fully enabled or disabled to allow (or block) light to pass through and strike the detector Can be controlled programmatically using the compressed sensing matrix A. Herein, it is assumed that the compressed sensing matrix A is a predetermined maximum length matrix. An example of such a matrix is a randomly or pseudo-randomly generated MxN Walsh-Hadamard matrix. In other embodiments, the matrix A is of any type suitable for use in compressed sensing, having some well-known properties in compressed sensing theory (eg, random row selection of orthogonal matrices). Can be a sparse matrix.

The compressed sensing camera device 106 is configured to provide the compressed basis vectors a ₁ , a ₂ ,. . . a Using each of _M , the respective times t ₁ , t ₂ ,. . . The output of the photon detector is periodically processed (eg, integrated, filtered, digitized) to generate a set of M compressed measurements y _k (kε [1... M]) over t _M Etc.). As will be further appreciated, the compression measurements y _k (kε [1... M]) collectively represent the compressed image of the scene 104. In practice, the M compression measurements that are generated are between the desired level of compression and the desired native resolution of a full resolution N-pixel image that can be reconstructed using the M compression measurements. Represents the predetermined equilibrium. The compressed sensing device 106 may be configured based on such balance.

A compressed N-pixel image x ₁ , x ₂ ,. . . Compression measurements represent the _{x N y 1, y 2,} y 3,. . . vector y y _M may be sent to the multi-resolution reconstruction device 110 via the network 108 by the compression sensing device 106.

  As noted above, in accordance with various aspects of the present disclosure, the reconstruction device 110 is configured to generate a multi-resolution image from the compressed measurements. In particular, and as described in detail below, the reconstruction device 110 is configured to generate an image from compressed measurements, where portions of the image are at a higher resolution than other portions of the image. For example, the image generated by the reconstruction device 110 may be one or more regions of interest that are generated at a possible full resolution (ie, a fixed single native resolution based on which compression measurements were obtained). And at least one or more other regions that are generated at a lower resolution than that of the region of interest.

  Although the devices or units are shown separately in FIG. 1 for ease of understanding, in some embodiments, the devices may be combined into a single unit or device. For example, in one embodiment, a single processing device may be configured to provide both the functionality of generating compressed measurements and reconstructing the desired image. A single processing device (as if the devices are separate) configured to provide memory to store one or more instructions and, when executed, the processor as described herein And a processor for executing one or more instructions. A single processing device typically includes computing, such as one or more input / output components for inputting or outputting information to / from the processing device, including cameras, displays, keyboards, mice, network adapters, etc. It may include other components found in the device. Network 108 may be an intranet, the Internet, or any type or combination of one or more wired or wireless networks.

  Next, the detailed operation of the reconstruction device 110 for generating a multi-resolution image from the compressed sample is described in conjunction with the process 200 shown in FIG.

In step 202, the reconstruction device 110 uses a compressed measurement y () that represents a compressed version of the single native resolution N-pixel image x of the scene generated by the compressed sensing device 106 using the MxN compressed sensing matrix. y ₁ , y ₂ ,... y _M ) are received or retrieved. In addition to the compressed measurements, in one embodiment, the reconstruction device 110 also receives from the compressed sensing device 106 the MxN compressed sensing matrix A, or MxN compressed sensing used by the compressed sensing device to generate compressed measurements. Information for generating the matrix A is received. For example, the sensing matrix may be generated by the reconstruction device based on the received seed value used by the compressed sense imaging device to generate the sensing matrix. In other embodiments, the compressed sensing device 106 used by the compressed sensing device 106 to obtain a compressed measurement generally does not change, so the compressed sensing device 106 may be in, for example, initialization of the compressed sensing device, etc. Prior to obtaining the compressed measurements, the compressed sensing matrix A may be sent to the reconstruction unit. In still other embodiments, the reconstruction matrix can be known in advance or can be generated by the reconstruction device 110.

  In step 204, the reconstruction device 110 determines the desired two-dimensional multi-resolution image size to be reconstructed from the compressed sense measurements. The reconstruction device 110 uses the magnitude as the row and column size corresponding to the row and column size of the two-dimensional representation of the one-dimensional N-pixel image x for which the compressed measurements have been generated using the sensing matrix. Where the number of rows times the number of columns is equal to N. One skilled in the art will understand how to represent a two-dimensional image as a one-dimensional vector and vice versa. For example, when generating a compressed sense matrix and compressed measurements, a predetermined concatenation of rows may be used to map a two-dimensional native resolution image to a one-dimensional vector x in the compressed sense imaging device, and the reconstruction device May use the same predetermined concatenation method to determine the desired two-dimensional multi-resolution image size corresponding to the size of the two-dimensional native resolution image. In other embodiments, the size of the two-dimensional native resolution image is known in advance or may be predetermined.

  In step 206, the reconstruction unit allocates the plurality of regions to the desired multi-resolution 2D image to be reconstructed. In other words, this can also be understood as dividing a two-dimensional image into a plurality of regions. Allocating multiple regions includes allocating at least one region of interest (ROI) that is considered more important than other regions in terms of resolution.

FIG. 3 shows a specific example of allocating four regions to the two-dimensional single native resolution image 302. As shown therein, a two-dimensional image is partitioned, divided, or allocated into four regions 304, 306, 308, and 310, where for purposes of example x ₁ The regions denoted as are assumed to be allocated regions of interest compared to the regions denoted as x ₂ , x ₃ , x ₄ which are lower resolution regions that are not part of the ROI. More generally, the region allocation is x = x ₁ +. . . + Be represented as x _P, where (in the example 4) P is the number of regions.

In step 208, each of a plurality of regions (also referred to as blocks) to be allocated is assigned to the desired multi-resolution image as a corresponding region having the desired respective resolution, where the desired multi-resolution image The respective resolution assigned to each of the regions of is less than or equal to the single native resolution of the compressed image represented by the compression measurement y (y ₁ , y ₂ ,... Y _M ) .

FIG. 4 shows step 208 for the particular example of FIG. 3 in which four regions are allocated. As seen in FIG. 4, image 402 is a two-dimensional single native resolution image represented by the extracted compressed measurements, and the fine grid in image 402 illustrates a single native resolution herein. Used to do. Image 404 is the desired multi-resolution image to be constructed from the compressed measurements. Image 404 shows the corresponding four regions x ₁ , x ₂ , x ₃ , and x ₄ allocated according to the previous step. In addition, a grid is shown within each of the four regions of the image 404 to show the respective resolution allocation to each of the four blocks.

From the indicated grid image 404, region x ₁ defined as a region of interest, is allocated a resolution equal to the native resolution, can be seen to have the highest resolution as compared to other regions. As the grid is also shown, each of the other regions x ₂ , x ₃ and x ₄ each has a different resolution (in descending order) that is all smaller than the native resolution (and smaller than the resolution of the determined region of interest) Allocated).

More generally, a single native resolution for a given one of P regions may be expressed as | x _i | = N _i , where the resolution of each of the regions is Can be allocated.

は、所望の解像度を有する所望のマルチ解像度画像

の領域ｉであり、ｑ_ｉは、ステップ２０６において単一ネイティブ解像度画像に割り振られるＰ個の領域における各領域ｘ_ｉについての各大きさ（水平および垂直）のためのダウンサンプリング係数である。 here,

The desired multi-resolution image with the desired resolution

Region i, and q _i is the downsampling factor for each size (horizontal and vertical) for each region x _i in the P regions assigned to the single native resolution image in step 206.

として表され、ここで、ｘ_ｉは、単一ネイティブ解像度画像中の所与の割り振られた領域であり、

は、所望のマルチ解像度画像中の対応する領域であり、ここで、Ｅ_ｉは所与の領域についての拡張行列である。図５は、図３および図４に示された領域ｘ_４の特定の例に関してこのステップを示す。ステップ２１０の結果として、単一ネイティブ解像度画像ｘ全体と、拡張行列Ｅ_１、Ｅ_２．．．．Ｅ_ｐと、所望のマルチ解像度画像の割り振られた解像度をもつ領域

との間の関係は、

によって与えられる。たとえば、領域

が、ネイティブ解像度と同じである解像度を有するように割り振られた場合、対応する拡張行列Ｅ_１は、

を満たす恒等行列であろう。 In step 210, the reconstruction device 110 defines an expansion matrix for each respective allocated region to perform a mapping between the multi-resolution region and the single native resolution image. In general, for each region, the expansion matrix is

Where x _i is a given allocated region in a single native resolution image;

Is the corresponding region in the desired multi-resolution image, where E _i is the expansion matrix for the given region. Figure 5 shows the steps with respect to specific examples of regions x ₄ shown in FIGS. As a result of step 210, the entire single native resolution image x and the expansion matrices E ₁ , E ₂ . . . . E _p and the area with the assigned resolution of the desired multi-resolution image

The relationship between

Given by. For example, the area

Is assigned to have a resolution that is the same as the native resolution, the corresponding expansion matrix E ₁ is

An identity matrix that satisfies

  In step 212, the reconstruction unit generates a multi-resolution reconstructed image of the scene using the allocated region, the expansion matrix, the received compressed sense measurements, and the compressed sensing matrix.

に変更される。変更は、所望のマルチ解像度画像のためのブロックの各々を再構築するために使用される。本質的に、このステップは、次式を反復的に解くことを伴う。

In this step, the general reconstruction step to generate the desired image from the compressed measurements is from y = Ax

Changed to The change is used to reconstruct each of the blocks for the desired multi-resolution image. In essence, this step involves iteratively solving:

は、マルチ解像度画像の再構築された領域のうちの１つである。α_ｉ、ｉ＝１、．．．、Ｐは、領域ｘ_ｉ、ｉ＝１、．．．、Ｐの面積を考慮に入れるためのスケーリング係数であり、ｆ_ｉ、ｉ＝１、．．．、Ｐは、ブロックの構造を促進する正規化関数（たとえば、Ｌ１ノルム）である。

は、再構築されるべき所望の解像度を有するマルチ解像度画像の割り振られた領域である。ｇは、測定値と獲得モデルとの間の距離を特徴づける関数（たとえば、Ｌ２ノルム）である。ｙは、圧縮測定値の受信されたセットである。Ａは、ｙを生成するために使用されたセンシング行列である。Ｅ_ｉは、所望の解像度を有する割り振られた領域

についての定義された拡張行列である。所望のマルチ解像度画像の領域

の各々が、解決され、取得されると、所望のフルまたは全マルチ解像度画像ｘ^＊が、

として取得され得る。シーンの生成されたマルチ解像度再構築画像は、シーンのマルチ解像度再構築画像として２次元形式でディスプレイデバイス上に表示され得る。 In the above formula, each

Is one of the reconstructed regions of the multi-resolution image. α _i , i = 1,. . . , P is a region x _i , i = 1,. . . , P is a scaling factor that takes into account the area of P, f _i , i = 1,. . . , P is a normalization function (eg, L1 norm) that promotes the structure of the block.

Is the allocated area of the multi-resolution image having the desired resolution to be reconstructed. g is a function that characterizes the distance between the measured value and the acquired model (eg, L2 norm). y is the received set of compression measurements. A is the sensing matrix used to generate y. E _i is the allocated area with the desired resolution

Is a defined extension matrix for. Desired multi-resolution image area

Each is resolved and acquired, the desired full or full multi-resolution image x ^*

Can be obtained as The generated multi-resolution reconstructed image of the scene can be displayed on the display device in a two-dimensional format as a multi-resolution reconstructed image of the scene.

  The systems and methods disclosed herein regenerate a multi-resolution image suitable for display on a display device based on processing a set of compressed measurements representing a compressed version of a single native resolution image. Allows construction or generation. In view of the present disclosure, it is believed that several advantages arise compared to more conventional approaches. The process presented herein allows for the reconstruction of multi-resolution images based on a set of compression measurements that represent a compressed version of a single native resolution image. Thus, the present disclosure provides that different images of different desired resolutions can be constructed from the same set of compression measurements without having to recapture or regenerate new compression measurements to support different desired resolutions. to enable. In addition, reconstruction of multi-resolution images occurs based on and using the same compression measurement that represents a compressed version of the native image, so reconstructed images at full native resolution are also desired if desired Can be constructed without the need to recapture the scene to generate new compression measurements for different resolutions.

  In addition, the region (s) of interest, the number of regions (and the resolution assigned to the regions) can be changed after the compression measurements are generated, and new compression measurements suitable for such changes. Different images can be reconstructed at different resolutions without having to generate Furthermore, this also means that the process for reconstructing multi-resolution images can occur in devices other than compressed sense image capture devices, so that the image capture device can sense based on different desired resolutions. Reconstruct the matrix and alleviate most of the computational resources that would be needed if the image had to be recaptured. Advantageously, this is a compressed sense image capture device without having to take into account the different regions or different resolutions that may be desired during processing and without having to reconfigure the compressed sense imaging device differently for different situations. Cost of the device and making the device suitable for mass deployment.

  Regions described herein, including regions of interest having higher resolution than other regions, can be determined in several ways. For example, in one embodiment, the region of interest (and other regions) may be determined based on manual input to the reconstruction device. In some embodiments, a very low resolution preview image may be constructed to identify one or more regions that include the region of interest, and then the desired multi-resolution image is constructed based on the preview image. obtain. In another embodiment, the region of interest can be automatically determined based on known or determined characteristics of the captured scene (eg, a door in a security application), etc., and the rest of the image is Can be considered a lower resolution region. In yet other embodiments, the region of interest may be determined such that the central area of the image is allocated with a higher resolution and then allocated as another region with a smaller resolution (or the same resolution) in descending order. One or more surrounding areas follow. As will be appreciated by those skilled in the art, to determine an area of interest, such as determining an overlapping coverage area based on multiple camera views capturing a scene and determining the overlapping area as an area of interest Other conventional approaches can also be used. Although four regions are shown in the above example, it will be appreciated that there can be a smaller or larger number of regions, each with an assigned resolution. In one embodiment, there may be one region of interest having a higher resolution (eg, native resolution), and the remainder of the image may be considered a second region having a lower resolution than the region of interest.

と元のネイティブ解像度領域ｘ_ｉとの間のマッピングを可能にする。一般的に言えば、元の解像度行列または領域（ｄ）を取得することは、サイズｑｘｑのブロックにおいて低解像度行列または領域

の要素を繰り返すことを含み、ここで、

である。特定の例として、４要素低解像度行列または領域

が、次式のように、拡張行列１_２ｘ２を使用してネイティブ解像度１６要素行列または領域ｄにマッピングされ得る。

The extended matrix (E) disclosed herein for each region is the respective lower resolution region.

And allows the mapping between the original native resolution area x _i. Generally speaking, obtaining the original resolution matrix or region (d) is a low resolution matrix or region in a block of size qxq.

Including repeating the elements of

It is. As a specific example, a four-element low-resolution matrix or region

_Can be mapped to a native resolution 16 element matrix or region d using the extension matrix _12x2 as follows:

を取得することは、次式のように、サイズｑｘｑのブロック（Ｂ_ｑｘｑ）の内部にある元の解像度行列（ｄ）の要素を合計することを伴う。

Going the other way, ie from the original native resolution region (d) to the low resolution matrix or region

Is accompanied by summing the elements of the original resolution matrix (d) inside the block (B _qxq ) of size qxq as follows:

  FIG. 6 illustrates a high-level block diagram of a computing device 600 suitable for implementing various aspects of the present disclosure (eg, one or more steps of process 200). Although shown in a single block, in other embodiments, apparatus 600 may also be implemented using parallel and distributed architectures. Thus, for example, one or more of the various units of architecture 100 of FIG. 1 described above, and other components disclosed herein, may be implemented using apparatus 200. Further, various steps, such as those shown in the example process 200, may be performed using the apparatus 600 sequentially, in parallel, or in a different order based on a particular implementation. Exemplary apparatus 600 includes a processor 602 (eg, a central processing unit (“CPU”)) communicatively interconnected with various input / output devices 604 and memory 606.

  The processor 602 may be any type of processor, such as a general purpose central processing unit (“CPU”) or a dedicated microprocessor such as an embedded microcontroller or digital signal processor (“DSP”). The input / output device 604 operates under the control of the processor 602 and inputs data to the device 600, such as, for example, network adapters, data ports, and various user interface devices such as a keyboard, keypad, mouse or display, or It can then be any peripheral device that is configured to output data.

  Memory 606 stores electronic information such as non-transitory memory such as temporary random access memory (RAM) or read-only memory (ROM), hard disk drive memory, database memory, compact disk drive memory, optical memory, It can be any type of memory suitable for accessing it, or a combination thereof. Memory 606 includes data and instructions that, when executed by processor 602, perform or perform the functions or aspects described above (eg, one or more steps of process 200). Configure the device 600 or have the device 600 implement or execute them. Further, apparatus 600 is stored in memory 606 and executed by processor 602 or other network protocols or other commonly found in computing systems such as operating systems, queue managers, device drivers, database drivers, etc. Components can also be included.

  Although a particular embodiment of apparatus 600 is shown in FIG. 6, various aspects in accordance with the present disclosure may be one or more application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), dedicated or programmable. Any other combination of hardware can also be implemented.

  Although aspects herein have been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the present disclosure. Accordingly, it should be understood that numerous modifications can be made to the exemplary embodiments and other configurations can be devised without departing from the spirit and scope of the disclosure.

Claims (10)

An apparatus for processing a compression sensed image comprising:
A memory for storing instructions and data;
A processor communicatively coupled to the memory and configured to access data and instructions, wherein the instructions are executed by the processor;
Retrieving M compressed sense measurements generated using an MxN compressed sensing matrix, wherein the compressed sense measurements represent a compressed version of a single native resolution N-pixel image of a scene And
Determining a desired two-dimensional multi-resolution image size to be reconstructed from the compressed sense measurements;
Allocating a plurality of regions to a desired two-dimensional multi-resolution image based on the determined size, wherein each of the allocated regions is equal to or equal to a native resolution of a single native resolution image Having a determined pixel resolution smaller than,
Defining an expansion matrix for each respective allocated region, the size of each expansion matrix being determined for each respective allocated region based on the determined pixel resolution of the allocated region Defining and
Configuring the processor to use a compressed sense measurement, a compressed sensing matrix, and a defined enhancement matrix to generate a multi-resolution reconstructed image of the scene;
apparatus.   The apparatus of claim 1, wherein the processor is further configured to allocate a plurality of regions after the compression measurements are retrieved.   The apparatus of claim 1, wherein the processor is further configured to allocate at least one region of interest having a determined pixel resolution higher than other allocated regions.   The processor is further configured to reconstruct the allocated region of interest by changing one of the allocated regions of interest or by adding additional regions of interest. The apparatus according to claim 3.   The apparatus of claim 3, wherein the processor is further configured to generate at least one region of interest based on manual input.   The apparatus of claim 3, wherein the processor is further configured to automatically generate at least one region of interest. A computer implemented method for processing a compression sensed image comprising:
Retrieving M compressed sense measurements generated using an MxN compressed sensing matrix, wherein the compressed sense measurements represent a compressed version of a single native resolution N-pixel image of a scene. When,
Determining a desired two-dimensional multi-resolution image size to be reconstructed from the compressed sense measurements;
Allocating a plurality of regions to a desired two-dimensional multi-resolution image based on the determined size, wherein each of the allocated regions is equal to or equal to a native resolution of a single native resolution image Allocating having a determined pixel resolution less than
Defining an expansion matrix for each respective allocated region, wherein the size of each expansion matrix is determined for each respective allocated region based on the determined pixel resolution of the allocated region Define the steps,
A computer-implemented method comprising generating a multi-resolution reconstructed image of a scene using compressed sense measurements, a compressed sensing matrix, and a defined enhancement matrix.   The computer-implemented method of claim 7, further comprising allocating a plurality of regions after the compression measurement is retrieved.   8. The computer-implemented method of claim 7, allocating at least one region of interest having a determined pixel resolution that is higher than other allocated regions.   10. Reconstructing the allocated region of interest by changing one of the allocated regions of interest or by adding additional regions of interest. A computer-implemented method according to claim 1.

Images